Multi-direction hand movement classification using EEG-based source space analysis.

Common spatial pattern (CSP) is a powerful algorithm for extracting discriminative brain patterns in motor imagery-based brain-computer interfaces (BCIs). However, its performance depends largely on the subject-specific frequency band and time segment. Accurate selection of most responsive frequency band and time segment remains a crucial problem. A novel evolutionary algorithm, the backtracking search optimization algorithm is used to find the optimal frequency band and the optimal combination of frequency band and time segment. The former is searched by a frequency window with changing width of which starting and ending points are selected by the backtracking optimization algorithm; the latter is searched by the same frequency window and an additional time window with fixed width. The three parameters, the starting and ending points of frequency window and the starting point of time window, are jointly optimized by the backtracking search optimization algorithm. Based on the chosen frequency band and fixed or chosen time segment, the same feature extraction is conducted by CSP and subsequent classification is carried out by Fisher discriminant analysis. The classification error rate is used as the objective function of the backtracking search optimization algorithm. The two methods, named BSA-F CSP and BSA-FT CSP, were evaluated on data set of BCI competition and compared with traditional wideband (8-30[Formula: see text]Hz) CSP. The classification results showed that backtracking search optimization algorithm can find much effective frequency band for EEG preprocessing compared to traditional broadband, substantially enhancing CSP performance in terms of classification accuracy. On the other hand, the backtracking search optimization algorithm for joint selection of frequency band and time segment can find their optimal combination, and thus can further improve classification rates.

Electroencephalogram (EEG) signals are an essential tool for the detection of epilepsy. Because of the complex time series and frequency features of EEG signals, traditional feature extraction methods have difficulty meeting the requirements of recognition performance. The tunable Q-factor wavelet transform (TQWT), which is a constant-Q transform that is easily invertible and modestly oversampled, has been successfully used for feature extraction of EEG signals. Because the constant-Q is set in advance and cannot be optimized, further applications of the TQWT are restricted. To solve this problem, the revised tunable Q-factor wavelet transform (RTQWT) is proposed in this paper. RTQWT is based on the weighted normalized entropy and overcomes the problems of a nontunable Q-factor and the lack of an optimized tunable criterion. In contrast to the continuous wavelet transform and the raw tunable Q-factor wavelet transform, the wavelet transform corresponding to the revised Q-factor, i.e., RTQWT, is sufficiently better adapted to the nonstationary nature of EEG signals. Therefore, the precise and specific characteristic subspaces obtained can improve the classification accuracy of EEG signals. The classification of the extracted features was performed using the decision tree, linear discriminant, naive Bayes, SVM and KNN classifiers. The performance of the new approach was tested by evaluating the accuracies of five time-frequency distributions: FT, EMD, DWT, CWT and TQWT. The experiments showed that the RTQWT proposed in this paper can be used to extract detailed features more effectively and improve the classification accuracy of EEG signals.

Motor imagery (MI) is the most used paradigm in Brain-Computer Interface (BCI). As in other BCI paradigms, it is divided into feature extraction and classification. In this study, we implemented two feature extraction methods, one based on the well-known Common Spatial Pattern (CSP) algorithm and the other based on a not so well-known electroencephalogram (EEG) source localization technique called Low-Resolution Brain Electromagnetic Tomography (LORETA). We evaluated 30 right-handed subjects from an EEG data set made publicly available through Giga Science, where participants performed MI of left- and right-hand movements. After feature extraction with CSP and LORETA, MI tasks classification were carried out using the Linear Discriminant Analysis (LDA), Support Vector Machines (SVM), Naïve Bayes (NB) and Multi-Layer Perceptron (MLP) algorithms. Finally, we evaluated classification performance with all possible combination of classifiers and feature extraction methods. For all classifiers, the CSP feature extraction method performed better than LORETA. The best classification accuracy for the LORETA method was 71.2% and for the CSP method was 94.2%, both achieved with SVM.

Current Brain Computer Interface (BCI) systems are still not commercialized due to their insufficient accuracies. Although Common Spatial Pattern (CSP) is an effective method for BCI applications, it does not consider the channels' co-activation information. This study is aimed at modifying CSP method by incorporating synchronization of channels in the CSP formula and then passes the values through a nonlinear kernel function to enhance the discrimination rate between the left and right imagery movements. Inserting the synchronization of every two channels into the kernel CSP formula is performed in two different ways. Experimental results on a set of EEG signals belonged to three subjects for discrimination of left and right hand imagery movements reveal that the proposed kernel synchronous CSP method outperformed the standard CSP, synchronous CSP and kernel CSP methods for all subjects.

Non-invasive Brain Computer Interface (BCI) using EEG or MEG is one of hot research areas nowadays. This technology gives a new way to communicate between human and computer or machine. BCI system enables persons to command a computer without spontaneous limb movement by measuring the brain signal and interpreting the intention of users in real-time basis. MEG/EEG signal at sensor should be combination of signals originated from infinitely many sources. So, signals at sensor are so noisy and are hard to discriminate different state information (feature), which contains implicitly user’s intention. In general, it is too difficult to improve the performance of EEG based BCI without seeking better feature extraction. For seeking different features from the existing features, we try to do source localization in the belief that projection of EEG data into source space enables us to see unseen information in sensor space. Eigenbased minimum variance beamformer is introduced for this purpose in this work. It gives reconstructed source image on each voxel and is too robust to noise as well as fast (real-time reconstruction is tractable in some sense), which is well-suited to BCI system. Each trial for the given condition is projected into source space by beamformer localizer. Then in the source space feature extractions and classification are attempted.

Multimodal collaborative BCI system based on the improved CSP feature extraction algorithm

Brain Computer Interface (BCI) based on motor imagery can translate the subject's EEG, which is captured from the scalp when they are imaging the movements of their limb, into a series of control signals. The patients suffered from locked in syndrome can use this BCI system communicates with the world. Due to the characters of non-linear and non-stationary with the human EEG, how to extract the valuable features from different EEG data based on motor imagery, will be the key problem to design an efficient BCI system. In this paper, a novel method named Hilbert Huang transform (HHT) is used to extract the features from different EEG data based on motor imagery. Genetic algorithm (GA) is used to select the most valuable features to release the pressure of the classifier for higher accuracy and faster speed. Compared with traditional frequency feature extraction method, HHT and GA gain much higher classification accuracy.

The purpose of this study is to motivate the use of the simpler Linear Discriminant (LD) classifier as compared to the commonly used Multilayer-perceptron-backpropagation (MLP-BP) neural network for Brain Computer Interface (BCI) design. We investigated the performances of MLP-BP and LD classifiers for mental task based BCI design. In the experimental study, EEG signals from five mental tasks were recorded from four subjects and the classification performances of different combinations of two mental tasks were studied for each subject. Two different AR models were used to compute the features from the electroencephalogram signals: Burg's algorithm (ARB) and Least Square algorithm (ARLS). The results showed that in most cases, LD classifier gave superior classification performance as compared to MLP-BP, with reduced computational complexity. However, the best mental tasks for each subject were the same using both classifiers. ARLS gave the best performance (93.10%) using MLP-BP and (97.00%) using LD. As the best mental task combinations varied between subjects, we conclude that for different subjects, proper selection of mental tasks and feature extraction methods would be essential for a BCI design.

One problem in the development of brain-computer interface (BCI) systems is to minimize the amount of subject training on the premise of accurate classification. Hence, the challenge is how to train the BCI system effectively especially in the scenario with small amount of training data. In this paper, we introduce improved semisupervised adaptation based on common spatial pattern (CSP) features. The feature extraction and classification are performed jointly and iteratively. In the iteration step, training data are expanded by part of the testing data with labels which are predicted by a linear discriminant analysis classifier and/or a Bayesian linear discriminant analysis classifier in the previous iteration. Then CSP features are reextracted from the expanded training data, and the classifiers are retrained. Both self-training and cotraining paradigms are proposed for the improved semisupervised adaptation. Throughout the investigation on different number of initial training trials, we find that when a small number of training trials are used, e.g., a training session contains no more than 30 trials, similar classification performance to that of large training data items (40-50 trials) can be achieved. Effectiveness of the algorithms is verified by two competition datasets. Compared with several existing algorithms, the proposed semisupervised algorithms show improvements in classification accuracy for most of the competition datasets especially in the case of small training data.

Feature extraction in electronic nose (E-nose) signals is the first signal processing step in machine olfactory systems. Traditional feature extraction methods are often cumbersome and depend greatly on the experience of the researcher. The failure of feature extraction seriously deteriorates the performance of subsequent pattern recognition algorithms. In addition, inherent sensor signal shifts among sensing systems with identical configurations affect E-nose performance. Therefore, due to the unfavorable performance of empirical feature extraction and poor reproducibility, the odor detection ability of E-nose systems has deteriorated. In this paper, a bionic olfactory signal processing (BOSP) network based on a spiking neural network (SNN) is proposed to address these issues. Inspired by the structure of the olfactory bulb in mammals, the system has two components: a virtual olfactory receptor (VOR) layer and a bionic olfactory bulb (BOB) layer. The VORs convert the sensor response signals into spike sequences through two spike coding strategies. The BOBs, consisting of mitral cells and granule cells, extract spatiotemporal features from the spike sequences. The proposed BOSP network is a feature-level model that automatically extracts useful features from the sensor array outputs without applying traditional and tedious feature extraction steps, which can subsequently cascade various classifiers. The results of extensive experiments demonstrate that the BOSP network not only shows better feature extraction capability than several state-of-the-art methods but also exhibits promising shift suppression ability.

This study describes a method for classifying electrocorticograms (ECoGs) based on motor imagery (MI) on the brain–computer interface (BCI) system. This method is different from the traditional feature extraction and classification method. In this paper, the proposed method employs the deep learning algorithm for extracting features and the traditional algorithm for classification. Specifically, we mainly use the convolution neural network (CNN) to extract the features from the training data and then classify those features by combing with the gradient boosting (GB) algorithm. The comprehensive study with CNN and GB algorithms will profoundly help us to obtain more feature information from brain activities, enabling us to obtain the classification results from human body actions. The performance of the proposed framework has been evaluated on the dataset I of BCI Competition III. Furthermore, the combination of deep learning and traditional algorithms provides some ideas for future research with the BCI systems.

Effective common spatial pattern (CSP) feature extraction for motor-imagery (MI) EEG recordings usually depends on the filter band selection to a large extent. However, the most proper filter band can hardly be determined manually due to its subject-specific property. This study introduces a sparse support vector machine (SSVM) approach to implement simultaneous CSP feature selection and classification for MI-based brain–computer interface (BCI). In SSVM, CSP features are first extracted on multiple signals that are filtered from raw EEG data at a set of overlapping subbands. SVM with l 1-norm regularization is then proposed to classify MI tasks with automatic selection of filter bands giving the significant CSP features. The effectiveness of SSVM for MI classification is demonstrated on the BCI Competition III dataset IVa, in comparison with several other competing methods. Experimental results indicate that the proposed SSVM method is promising for development of an improved MI-based BCI.

Common spatial pattern (CSP) is a popular algorithm for spatial filtering and subsequent feature extraction in motor imagery-based (MI) brain-computer interface (BCI) systems. The performance of CSP, however, depends heavily on the subject-specific frequency band and time segment used for classifying mental tasks. Accurate selection of most informative frequency band and time segment poses a great challenge. In this study, quantum particle swarm optimization is proposed for sole selection of frequency band and joint selection of frequency band and time segment, which are realized by a wrapping approach, incorporating CSP for feature extraction and support vector machine for classification into the classification model. The classification error rate is used as the fitness function of quantum particle swarm optimization. The classification performance of quantum particle swarm optimization based CSP algorithm for joint selection of frequency band and time segment is evaluated by comparing with other three CSP algorithms using either fixed frequency band and time segment or fixed time segment and the frequency band selected by particle swarm optimization and quantum-behaved particle swarm optimization, on two MI data sets with different number of channels and trials. Experimental results suggest that the proposed algorithm outperforms the other three algorithms in terms of classification error rate. Across all subjects from the two data sets, the averaged error rate of the proposed algorithm was 7.45%, 2.97% and 2.05% lower than the CSP with fixed frequency band and time segment, that with selected frequency bands by particle swarm optimization and that with selected bands by quantum particle swarm optimization. The proposed algorithm can facilitate the real-world application of BCIs.

The usage of methods for the estimation of the true underlying connectivity among the observed variables of a system is increasing, especially in the domain of neuroscience. Granger causality and similar concepts are employed for the estimation of the brain network from electroencephalogram (EEG) data. Also source localization techniques, such as the standardized low resolution electromagnetic tomography (sLORETA), are widely used for obtaining more reliable data in the source space. In this work, connectivity structures are estimated in the sensor and in the source space making use of the sLORETA transformation for simulated and for EEG data with episodes of spontaneous epileptiform discharges (ED). From the comparative simulation study on high-dimensional coupled stochastic and deterministic systems originating in the sensor space, we conclude that the structure of the estimated causality networks differs in the sensor space and in the source space. Moreover, different network types, such as random, small-world and scale-free, can be better discriminated on the basis of the data in the original sensor space than on the transformed data in the source space. Similarly, in EEG epochs containing epileptiform discharges, the discriminative ability of network topological indices was significantly better in the sensor compared to the source level. In conclusion, causality networks constructed at the sensor and source level, for both simulated and empirical data, exhibit significant structural differences. These observations indicate that further studies are warranted in order to clarify the exact relationship between data registered in the sensor and source space.

Brain-computer interface (BCI) systems are proposed as a means of communication for locked-in patients. One common BCI paradigm is motor imagery in which the user controls a BCI by imagining movements of different body parts. It is known that imagining different body parts results in event-related desynchronization (ERD) in various frequency bands. Existing methods such as common spatial patterns (CSP) and its refinement filterbank common spatial patterns (FB-CSP) aim at finding features that are informative for classification of the motor imagery class. Our proposed method is a temporally adaptive common spatial patterns implementation of the commonly used filter-bank common spatial patterns method using convolutional neural networks; hence it is called TA-CSPNN. With this method we aim to: (1) make the feature extraction and classification end-to-end, (2) base it on the way CSP/FBCSP extracts relevant features, and finally, (3) reduce the number of trainable parameters compared to existing deep learning methods to improve generalizability in noisy data such as EEG. More importantly, we show that this reduction in parameters does not affect performance and in fact the trained network generalizes better for data from some participants. We show our results on two datasets, one publicly available from BCI Competition IV, dataset 2a and another in-house motor imagery dataset.